Agentic AI in Logistics: Why 55% Accuracy Fails

Agentic AI models for logistics fail when spatial reasoning comes into play, according to Bart Coppelmans, head of enterprise product at HERE Technologies. Despite general AI models correctly answering about 55% of basic direction questions, they lack the necessary spatial reasoning required for real-world freight operations. Tasks like rerouting trucks around bridges, identifying parking spots, and anticipating congestion require dense webs of location-specific data that current large language models do not possess.

HERE Technologies addresses this challenge by embedding location intelligence at the foundation of their logistics AI, not adding it later. This approach creates a feedback loop that captures driver input and builds a learning pattern to prevent repeating the same failure modes. Coppelmans emphasizes that agentic AI should be seen as an assisted journey with humans remaining in control, allowing planners to make adjustments based on cost, compliance, risk, safety, and sustainability.

The company is also working towards multi-agent collaboration across customer and third-party systems to eliminate fragmentation and enable seamless integration across workflows.